AI Is Taking White-Collar Jobs. Here's Which Ones Are Already Gone

In 2025, companies blamed AI for 55,000 job cuts. That's more than 12 times what they blamed it for two years earlier. Goldman Sachs has been tracking the number quietly — AI is shaving roughly 16,000 jobs per month off US payroll growth. And that's not a projection. That's what's happening now.

The CEOs who built these tools are saying things in public that should probably alarm more people than they do.

The Jobs Already Going

Some roles are going faster than others. The common combo? Lots of volume, lots of repetition, lots of tasks you can describe in a sentence. Those are the ones going first.

Customer service. Klarna tried to replace 700 support workers with AI and spent two years telling everyone it worked. It didn't, fully. Customer satisfaction dropped badly on anything complicated — disputes, edge cases, situations where someone was upset. By early 2026 the CEO admitted they'd cut too deep and started hiring humans again. Klarna is now the story everyone in a boardroom has to answer for before they greenlight an AI replacement plan. The model that actually works, they've found, is AI handling the easy stuff and humans handling everything that requires judgment. That's probably not what anyone was hoping to hear after cutting half their workforce.

HR and back-office at IBM. IBM's CEO confirmed AI had replaced around 200 HR workers — part of a broader restructuring that saw the company pause hiring across thousands of back-office roles. Problem resolution times worsened, morale dropped, and IBM began rebuilding teams in areas requiring human contact. Not a full reversal — but a messy one.

Duolingo went "AI-first" and cut the contractors doing translation and content work. There was enough backlash that the CEO had to go on a clarification tour. They've since walked back the more aggressive parts.

Beyond those headlines: entry-level legal work, paralegal research, junior analyst roles at banks, data entry, admin coordinators. The World Economic Forum specifically named administrative assistants, bank tellers, and data entry clerks as seeing the largest absolute headcount decline of any job category through 2030. Not the largest percentage — the largest raw numbers.

The People Building This Are Saying Uncomfortable Things

Dario Amodei runs Anthropic. He told Axios last May that AI could wipe out half of all entry-level white-collar jobs within five years. He said US unemployment could hit 10–20%. He said the shift from AI-as-tool to AI-as-worker would happen "in as little as a couple of years or less."

This is the person who built the thing saying this. Not a pundit.

Mustafa Suleyman runs AI at Microsoft. In February he said that most work done "sitting down at a computer" — accounting, legal, marketing, project management — would be fully automated within 12 to 18 months.

You can disagree with both of them. A lot of smart people do. But it's worth noticing who's saying it.

What the Numbers Actually Say

Goldman Sachs has AI at around 16,000 jobs per month knocked off payroll growth. Not 16,000 fired — 16,000 fewer jobs added than there would have been. Unemployment among 20–30 year olds in tech-exposed roles is up 3 percentage points since early 2025.

McKinsey estimates roughly 57% of US work hours involve tasks today's AI could theoretically handle. Not 57% of jobs gone overnight — just that more than half of what people actually do at work could, technically, be done by a machine that already exists.

WEF says 92 million jobs go by 2030, 170 million new ones come. Net positive — 78 million. But those jobs won't be in the same companies, the same cities, or for the same people who lost the first ones. And 40% of employers told WEF they're planning to cut headcount wherever AI can take over tasks.

Jamie Dimon said at Davos that JPMorgan — 318,000 employees — will probably be smaller in five years even as the business grows. The CFO has flagged a potential 10% cut in operations driven by AI. That's not a tech startup optimizing its burn rate. That's the largest bank in the country.

It's Landing on Young People First

Workers aged 22 to 25 in AI-exposed roles have seen a 13% employment drop since 2022. Goldman Sachs put it plainly: Gen Z is taking the brunt. Not because they're less capable — because entry-level white-collar work is exactly what AI does first. The document reviews. The research pulls. The first drafts. The data summaries.

Wall Street has quietly communicated plans to cut around 200,000 jobs over the next few years, most of it at the entry and back-office levels.

Those entry roles weren't just jobs. They were how people learned the industry, built relationships, got promoted. For a generation already on track to be poorer than their parents, losing that on-ramp is a different kind of problem.

What's Actually Growing

Not everything is contracting. Some areas are genuinely taking off.

Cybersecurity is growing fast — 29% projected growth from 2024 to 2034, per the Bureau of Labor Statistics. AI creates more attack surface, which means more demand for the humans securing it. US job postings requiring AI skills were up 144% year-over-year as of April 2026. Healthcare administration is hiring because AI adoption there is slower by necessity — you can't automate your way through regulatory risk in medicine as fast as you can in a call center. Green energy and infrastructure roles are expanding as climate spending scales up.

The pattern that's emerging: the roles managing, building, and directing AI are growing. The roles AI directly replaces are shrinking. That's not a new story — it's the same one that played out with every wave of automation before this. It just feels different when it's office jobs.

What You Can Do With This

"Learn to code" keeps getting repeated as the answer. It's probably wrong. Software development is one of the most exposed fields — Amazon, Google, and Microsoft have all said their AI systems are handling a growing share of engineering output.

The wage data points somewhere more specific. PwC analyzed close to a billion job postings across six continents and found workers with real AI skills earn a 56% wage premium over colleagues in the same role without them. A year earlier that premium was 25%. It's more than doubled in 12 months. Lightcast puts it at 28% across industries — roughly $18,000 more per year.

The foreman didn't lose his job when the assembly line arrived. The workers doing the physical task did. The foreman understood the system, handled the exceptions, and made the calls a machine couldn't. That's still the answer — just for a different kind of machine.

The E-shaped economy has been building for years. Gains concentrate at the top, the middle hollows out, the bottom stagnates. AI didn't start that. But everything in the data suggests it's making it move faster.